At 01:51:44 Universal Time (ut) on 4 September 2005, Swift's Burst Alert Telescope (BAT) detected GRB 050904 and 81 seconds later a 4′-radius localization was distributed to observers on the ground. Swift's X-Ray Telescope (XRT) automatically slewed to the BAT localization and 76 minutes after the burst a 6″-radius XRT localization was distributed.9

Over the next few hours, we observed the XRT localization at both near-infrared (NIR) and visible wavelengths (Table 1). In the NIR, we discovered a bright (J ≈ 17.4 mag at 3.1 hours after the burst) and fading source within the XRT localization using the 4.1-m Southern Observatory for Astrophysical Research (SOAR) telescope on top of Cerro Pachon in Chile (Fig. 1).4

Table 1 Observations of the afterglow of GRB 050904
Figure 1: NIR and visible-light images of the field of GRB 050904.
figure 1

Left, NIR discovery image of the bright (J = 17.36 ± 0.04 mag) afterglow of GRB 050904 from 4.1-m SOAR on top of Cerro Pachon, Chile. Middle, near-simultaneous non-detection of the afterglow at visible wavelengths (unfiltered, calibrated to Rc > 20.1 mag) from one of the six 0.41-m PROMPT telescopes that we are building on top of Cerro Tololo, which is only 10 km away from Cerro Pachon. Right, colour composite (r′i′z′) image of the afterglow 3.2 days after the burst from 8.1-m Gemini South, which is also on top of Cerro Pachon.

However, at visible wavelengths we did not detect the afterglow to relatively deep limiting magnitudes using one of the six 0.41-m Panchromatic Robotic Optical Monitoring and Polarimetry Telescopes (PROMPT)10 that we are building on top of Cerro Tololo, which is only 10 km away from Cerro Pachon4, the 60-inch telescope at Palomar Observatory in California11, and the 3.5-m telescope at Calar Alto Observatory in Spain. Nor did we detect the afterglow with the 0.30-m Burst Observer and Optical Transient Exploring System (BOOTES)12 1B telescope in El Arenosillo, Spain, which began imaging the field only 2.1 minutes after the burst13. This implies that the GRB either occurred at a very high redshift or was very heavily extinguished by dust4.

Between about 3 hours and about 0.5 days after the burst, the fading of the afterglow appears to be well described by a power law of index (Fig. 2)5,6. However, after this time the fading appears to have slowed to a temporal index of (refs 7, 14, 15). A single power-law description is ruled out at the 3.7σ credible level. One possible explanation is that our initial SOAR observations caught the tail end of a reverse shock that had been stretched in time by a factor of 7.29 owing to cosmological time dilation. Even so, the reverse shock would still be at least a few times longer-lived in the source frame than the reverse shocks of GRBs 990123 and 021211 (refs 16, 17). Another possibility is that we are undersampling a light curve that is undergoing temporal variations, such as in the afterglows of GRBs 021004 and 030329 (refs 18, 19). Indeed, the X-ray afterglow is extremely variable at these times20.

Figure 2: NIR and z′-band light curves of the afterglow of GRB 050904 and our best-fit model. A single power-law description is ruled out at the 3.7σ credible level.
figure 2

Following the formalism of Frail et al.26, given GRB 050904's redshift and fluence27 the non-detection of a jet break in the light curve prior to 2.3 days after the burst implies that the opening/viewing angle of the jet is 3° and that the total energy that was released in γ rays is 5 × 1050 erg. The Z-band measurement (unfilled circle) is a factor of three below the fitted model, but this appears to be real and temporal in nature (Table 1). Error bars are s.e.m. Downward arrow indicates upper limit.

Using these temporal indices to scale all of our measurements to a common time, except for a Z-band (0.84–0.93 µm) measurement from 11.4 hours after the burst (Table 1), we plot the spectral flux distribution of the afterglow in Fig. 3. In the NIR, the spectral index is However, the spectral index between NIR and visible wavelengths is steeper than -5.9. This is too sharp a transition to be explained by dust extinction alone5,21. However, a small amount of extinction cannot be ruled out and is probably present (Fig. 3).

Figure 3: Spectral flux distribution of the afterglow of GRB 050904 and our best-fit model.
figure 3

Measurements have been scaled to 10.6 hours after the burst using our best-fit light curve. We model the spectrum as a power law with negligible emission blueward of Lyα. Shallower intrinsic power-law spectra can be inferred with the addition of source-frame dust. If one assumes that the jet is propagating through either a constant-density or wind-swept medium with the synchrotron electron cooling frequency either redward or blueward of the observed frequencies, the fitted temporal index implies an electron energy distribution index between and and an intrinsic spectral index between and (refs 28, 29). This is shallower than the fitted spectral index which suggests that source-frame dust is probably present. However, only a small amount is required to explain such a difference at these source-frame frequencies. This cannot explain the sharp drop in spectral flux in and blueward of the z′ band5,21. We take Galactic E(B - V) = 0.060 mag (ref. 30). Error bars are s.e.m. Downward arrows indicate upper limits.

Assuming negligible emission blueward of Lyα, we measure a photometric redshift of (refs 5, 6–7), which is consistent with the spectroscopic redshift of z = 6.29 ± 0.01 (ref. 8). For H0 = 71 km s-1 Mpc-1, Ωm = 0.27, and ΩΛ = 0.73 (ref. 22), this corresponds to about 900 million years after the Big Bang, when the Universe was about 6% of its current age. The next-most-distant GRB that has been identified occurred at z = 4.50 (ref. 3), which was about 500 million years later, when the Universe was about 10% of its current age.

One of the most exciting aspects of this discovery is the brightness of the afterglow: extrapolating back to a few minutes after the burst, the afterglow must have been exceptionally bright redward of Lyα for the robotic 0.25-m TAROT telescope to detect it in unfiltered visible-light observations23. Extrapolating our J-band light curve back to these times yields J ≈ 11–12 mag. This suggests that by pairing visible-light robotic telescopes with NIR robotic telescopes, and these with larger telescopes that are capable of quick-response NIR spectroscopy, all preferably at the same site so that they are subject to the same observing conditions, at least some very-high-redshift afterglows will be discovered, identified, and their NIR spectrum taken while they are still sufficiently bright to serve as a powerful probe of the conditions of the early Universe10.